Before AI, I rarely heard non-developers talk about building their own apps, tools, or automations. Now it’s everywhere, and this is largely due to “vibe-coding” tools, including Claude Code. The reason I’m specifically highlighting Claude Code is because Claude has become the tool that everyone (and I mean literally everyone) has been talking about lately.
Even users who would never touch a tool with the word “Code” in it set it up just to make sure they got their money’s worth with that $20 subscription. Now, I’m not going to be dramatic here and say that I don’t use Claude Code. Contrary to that, it’s one of my favorite tools, and I’ve been using it for months simply because it’s awesome. But somewhere along the way I started using OpenCode on the side, and then some more, and then I realized that I really wasn’t missing anything.
Want to stay up to date with the latest in AI? The XDA AI Insider newsletter is published weekly with in-depth reviews, tool recommendations, and practical coverage you won’t find anywhere else on the site. Subscribe by changing your newsletter preferences!
OpenCode doesn’t just lock you to Claude models
This isn’t surprising at all, but since Claude Code is an Anthropic product, you’re limited to Claude models. That’s it. No GPT, no Gemini, no local LLMs, nothing. You have access to Claude’s Haiku, Sonnet and Opus models. To be honest, they are excellent and that is why Claude Code is absolutely worth paying for. But it’s also the whole menu. If tomorrow Google comes up with a model that better suits your specific use case, or you want to run something locally for privacy reasons, you’re out of luck.
For example, GPT-5.5 was released last week and is significantly better for certain coding tasks. It obtained 82.7% on Terminal-Bench 2.0, well ahead of Claude Opus 4.7’s 69.4%. This is a benchmark that tests actual command line workflows, which is literally what these tools do. Now, Opus still wins over other things like SWE-Bench Pro, but that’s kind of the point! Different models are better for different tasks. With Claude Code, you don’t choose. With OpenCode, this is possible since the tool is strictly “bring your own model”. You get a very similar terminal interface to Claude Code, but you choose what runs underneath. OpenCode uses the AI SDK and Models.dev to support over 75 LLM providers, as well as local LLMs.
I gave Claude Code control of my office for a week, and he automated things I didn’t think possible
I was seriously stunned.
The best part is that you can also use some of your existing AI subscriptions with OpenCode. This includes ChatGPT Plus, GitHub Copilot and GitLab Duo. For the rest, you’ll have to tinker with the API keys, but it’s simple! You run a /connect command, your keys are stored locally, and you’re up and running. OpenCode also technically allows you to connect your Claude subscription to it, but Anthropic explicitly prohibits this now. However, you can still use Claude with OpenCode via an API key. You simply can no longer enjoy your Pro or Max subscription.
You can switch between models depending on the task. Use GPT-5.5 for terminal-intensive workflows where it performs better, switch to a Gemini model for fast iteration, and switch to a local model through Ollama for fast commit messages where you don’t need the big guns. Same tool, same interface, different brain behind it!
It also has many Claude Code-like features.
When I first installed OpenCode, I thought it would lack many of the features that made Claude Code…Claude Code. Turns out the majority of things I thought were Claude Code exclusives were already there. For example, a Claude Code feature that I find very useful in my coding sessions is Outline mode. Since I’m not a developer in the traditional sense, I rely heavily on Outline mode. Whether I’m making a change to an existing code base or building something from scratch, it’s very helpful for the model to analyze and map things before it goes haywire and makes changes itself.
Outline mode is designed to do just that. It keeps the agent in read-only mode so it can explore your project structure, understand dependencies, and build a full-fledged strategy before touching anything. OpenCode has that too! You move on to Plan mode, get your analysis, then move on to Build mode when you are ready to start building. OpenCode is also fully capable of editing files across your entire project, running terminal commands, and iterating over errors without you having to hold its hand.
It supports MCP servers, so you can connect it to all your favorite tools. You can also create skills: sets of reusable instructions that your agent can load on demand for specific tasks. It works the same way as in Claude Code: you create a SKILL.md file, define what it does and the agent extracts it when relevant.
The advantage isn’t Claude Code, it’s Claude
What I have tried to ensure throughout this article is that this is not considered a success. I really like Claude Code and am subscribed to the Max 5x plan with no plans to cancel my subscription. However, the more I used OpenCode, the more I realized that what I liked was Claude himself, not Claude Code.
Intelligence, reasoning, code quality, it’s all Claude. Claude Code is essentially just a wrapper around itself, so why would you really want to lock yourself into one tool when the model works just as well in another? One that also supports over 75 other providers, local models, and your existing subscriptions.
And if you’re migrating from Claude Code, OpenCode natively reads your existing CLAUDE.md files and skills directory as a fallback, so you don’t need to start from scratch! The tool is also completely free and open source, so what do you really have to lose by trying?